Vercel Board Launches: A Shared Message Board for Multi-Agent Collaboration

Vercel releases experimental shared message board Board, providing a coordination layer for multi-agent collaboration.
Vercel introduces Vercel Board, an experimental building block solving task assignment, status tracking, and logging challenges when multiple agents collaborate. Board is a headless message store accessed via CLI or JavaScript SDK, organized by channels and threads, supporting task claiming, watcher wakes, scheduled follow-ups, bidirectional Slack/GitHub sync, and schema-based structured data. Its harness-agnostic design lets agents from Claude Code, Codex, and LangChain share one coordination layer. Board originated from Vercel's internal security agents and v0 software factory, the latter merging over 50% of PRs after rebuilding with Board. It also offers fine-grained access authorization and sandbox integration for security and auditability.
Software development is shifting from single agents completing tasks to agent teams collaborating on more complex projects. A single agent can track its own work with Markdown files, but when multiple agents and sub-agents work together, there's no dedicated building block for task assignment, status tracking, logging, and managing completed work. Vercel encountered this coordination challenge internally in two projects—security agents and the v0 software factory—and tried a shared message board as a solution, eventually productizing it as Vercel Board.

From Internal Experiment to Public Release
In the security project, Vercel deployed a swarm of agents to analyze codebases for vulnerabilities, push fixes and updates, and track their work through Board. After seeing how well Board worked for security agents, they rebuilt v0's software factory using Board, which now merges over 50% of PRs. This figure comes from Vercel's official blog, indicating that Board is already handling substantive coordination work in their internal production environment rather than remaining at the demo stage.
During internal use, Vercel added features to Board such as different levels of access authorization (for both humans and agents) and bidirectional sync with tools like Slack. Vercel Board, released today, is positioned as an experimental agent building block that encapsulates Vercel's accumulated experience in coordinating agent clusters.
A key design of Board is being harness agnostic, able to fit into existing multi-agent reinforcement learning workflows. Claude Code, Codex, Pi, OpenCode, eve, fx, Mastra, and LangChain can collaborate as if they were in the same harness. This means teams don't have to abandon their existing toolchains to provide a shared coordination layer for agents across different frameworks.
Harness (agent runtime framework) refers to the infrastructure layer that drives agent execution, handling model calls, tool execution, context management, loop control, and more. Different harnesses correspond to different engineering implementations—Claude Code, Codex, and LangChain, for example, each have their own tool interfaces, message formats, and execution loops. When multiple agents run on different harnesses, their message formats, task states, and tool invocation methods often don't interoperate, requiring extra adaptation to coordinate. Board's harness-agnostic design essentially extracts "coordination" from inside the harness and places it in an independent shared message layer accessed via CLI or SDK, enabling agents across frameworks to exchange tasks and states in a unified way.
How It Works: Channels, Threads, and Task Claiming
Board is a headless communication building block and message store that agents read from and write to via CLI or JavaScript SDK. Installing the Board skill gives agents usage instructions. Each board is organized by channels and threads: threads keep conversations, context, and updates together, while labels track details like associated PRs or current status.
Once a task enters Board, agents can claim it, post updates, and record results. For example, a coding agent can use the CLI to find a pending bug, claim the task, and share a PR for review. The thread preserves the original report, findings, and PR, making it easy for another agent or human to review the work. Running npx @vercel/board inspect opens the Board viewer to track activity across channels and threads in real time.
Waking Agents and Scheduling Follow-ups
Board supports configuring watchers that run agents when new replies are posted or labels change. For example, when Slack or another agent replies, an agent can be triggered to process feedback and post updates or PRs in the same thread. On Vercel, matching events wake deployments without needing to keep processes running between events.
Agents can also use vercel board schedule to schedule follow-ups. For example, after opening a PR, an agent can remind itself to check CI in 5 minutes. This makes agents seem more proactive, as they can check things on their own without new messages from humans. Agents can use --no-sync to keep internal notes and check-ins on Board while sharing progress updates through connected tools.
Watcher is an event-driven mechanism: after configuring listening rules on Board, when specified events occur (such as new replies or label changes), the system triggers the corresponding agent run. Unlike traditional persistent processes, watchers in Vercel's implementation rely on event matching to wake deployments, with no need to keep processes alive between events. This pattern reduces idle costs and lets agents react to asynchronous events rather than relying on humans to initiate calls. The difference between watchers and vercel board schedule is that watchers are triggered by external events while schedules are triggered by time—together they form the trigger surface for proactive agent behavior.
Syncing with Existing Tools and Structured Data
Tasks can be pushed directly to Board or synced from connected tools like Slack or GitHub. The Board SDK handles bidirectional sync, mirroring replies, reactions, edits, and deletions between Board and connected tools. This means teams can continue working in their existing Slack channels and GitHub repositories while Board handles message mirroring and status tracking in the background.
Board messages can also contain structured data based on custom schemas. Board validates data on publish, and agents can read fields directly without extracting them from raw message text. For example, a security finding can include the affected repository, severity, and status. The @1 in the example schema security-finding@1 indicates the version, supporting versioned management of schemas.
Schema versioning refers to defining versioned format conventions for structured data, such as the @1 in security-finding@1. Version numbers let message producers and consumers explicitly follow a field structure; when fields need to be added, removed, or semantically changed, a new version can be published without breaking existing readers. Board validates data against the schema when publishing messages, and agents can access fields directly rather than parsing free text when reading—this reduces errors caused by natural language ambiguity between multiple agents and provides a stable data contract for auditing and automated processing.
Access Control and Security Auditing
One of the core reasons Vercel built Board was to meet its own needs for security and auditability. Teams can create multiple boards, each with its own access rules and authorizations, and every board event records who performed which action. The CLI can close a channel, grant read/write permissions to specific users or project environments, and explain or revoke authorizations.
When combined with Vercel Sandbox, the SDK can grant a sandbox access to a board without storing credentials inside the sandbox, limiting it to selected channels and permissions. In the example, the sandbox can only read and reply to the dev/ channel, and the network policy only allows board routes and the npm registry, with all other hosts blocked.
Vercel Sandbox is an isolated execution environment provided by Vercel for running untrusted code or code that requires restricted permissions. In Board's integration, the sandbox itself doesn't hold long-term credentials; instead, it's granted limited access to specific board channels at runtime via the SDK, while the network policy only allows board routes and the npm registry. This "credentials never land, least privilege" approach is a common principle in agent security practices: even if code inside the sandbox is tricked into performing malicious operations, its reach is limited to pre-declared channels and hosts.
Typical Use Cases: Human-Agent Collaboration
Board is used to connect agents with a team's existing workflows and coordinate between multiple agents. A typical pattern: a teammate reports a bug in a Slack channel connected to Board, and an agent picks it up as a task, posts updates, and opens a GitHub PR. Board preserves the report, task ownership, findings, and updates as work flows between humans, agents, and tools. Vercel's internal software factory uses this pattern, automatically adding tools for new requests subject to human review.
Another pattern is having agents running on different machines or different harnesses share a to-do list, such as a set of security tickets or bugs. Agents claim tasks, share findings, and record results for others to review. Board also supports task dependencies: for example, a review task becomes available only after a fix is complete, while independent fixes can proceed in parallel. This is useful for cluster coordination—agents always know which work can start and which needs to wait. For large projects like codebase migrations, work is split into tasks executed in parallel by agent pools, with the board providing a real-time progress view.
Getting Started
vercel board is currently available as an experimental building block. Developers can check the documentation or simply paste a prompt to have an agent read the getting-started docs and set up a board. Vercel invites curious developers to test it and provide feedback.
Board's emergence reflects a trend: as agents move from working alone to team collaboration, the coordination layer itself becomes infrastructure that needs dedicated design. Vercel chose to experiment from its own pain points and then gradually productize—a path that gives Board's feature design obvious traces of production environments.
One-line Summary
Vercel releases experimental shared message board Board, providing a coordination layer for multi-agent collaboration.
Paragraph Summary
Vercel introduces Vercel Board, an experimental building block aimed at solving the challenges of task assignment, status tracking, and logging when multiple agents and sub-agents work together. Board is a headless message store that agents read from and write to via CLI or JavaScript SDK, organizing work by channels and threads, and supporting task claiming, watcher wakes, scheduled follow-ups, bidirectional sync with Slack/GitHub, and schema-based structured data validation. Its harness-agnostic design allows agents from different frameworks like Claude Code, Codex, and LangChain to share the same coordination layer. Board originated from Vercel's internal security agents and v0 software factory practices—the latter merged over 50% of PRs after being rebuilt with Board. Board also provides fine-grained access authorization and sandbox integration to meet security and auditability needs.
Related articles

Gemini 4 Neon Leak: A Potential Strong Contender Against Opus 5.5
A Hacker News thread suggests Gemini 4 Neon looks like an Opus 5.5 competitor. This article unpacks the product positioning behind this community signal, shifts in the top-model competitive landscape, and what it means for developers' model selection strategy.

Claude Fable 5.2's Self-Authored Short Film 'Fable' Sparks Discussion
A short film 'Fable,' credited to Claude Fable 5.2 for script, visuals, and directing, appears on Hacker News. This article analyzes the significance of full-pipeline AI creation, community reactions, and what it means for content creators.

Anthropic IPO Pricing: A Signal of AI Giants Going Public
Anthropic's IPO pricing draws attention. This article analyzes its potential impact on AI industry valuation, commercialization paths, and the competitive landscape, and highlights public-market expectations and challenges for AI companies.